ML DFT benchmarking

This dataset consists of 92 reaction profiles computed in a high-throughput manner at CAMB3LYP-GD3BJ/6-311++G** level of theory with the help of autodE and Gaussian 16 in the gas phase. Gaussian log-files for the IRC and the final frequency calculation for each reactant, TS, and product species, XYZ-files of final geometries as well as a CSV file containing computed electronic energies and thermal corrections are available in a compressed archive file, full_dataset_profiles.tar.gz.

The files have been organized per reaction profile, and identified through the reaction ID. Within each directory, reactant XYZ-files are of the form r_####.xyz, product XYZ-files are of the form p_####.xyz, and transition state XYZ-files are of the form TS_####.xyz. If the product/reactant had to be corrected to enforce stereochemical compatibility, the latter XYZ-files are included under the form of p/r_alt_####.xyz. The frequency log-files can be found in a frequency_logs directory. 

In the single_points.tar.gz directory,  log-files for the computation with the 20 diferent functionals can be found. 

The references DLPNO-CCSD(T) log-files calculations are available in the Reference values directory. 

If this dataset are used as part of a publication please cite the associated preprint.